CYBRARY Cybrary Forge Agent Range
Your AI assistants deserve a cyber range of their own.
Connect ChatGPT, Claude, Copilot or your own agents to Cybrary Forge over MCP. They practice incidents on a realistic simulated estate, your people work beside them, and every single action stays scoped, approved and audited.
Works with the assistants you already have
Three people care about this, for three different reasons
Everyone is being sold an autonomous SOC. Almost nobody can say what their agents actually do under pressure. Put yours on a range that behaves like your environment and find out before an incident does — with a scorecard you can show a board, an auditor or a customer.
Analysts practice the investigation, not the paperwork: realistic hosts, telemetry and consoles, with an AI colleague working the case beside them. Teams rehearse the handoffs, the escalation and the awkward 3am decision while it is still a drill.
Give an assistant a safe place to be genuinely useful. It can search the catalog, open a run, work a case and hand back evidence — inside a boundary you drew, where the worst possible outcome is a bad grade.
What makes it different
It plugs into what you already use
The range speaks Model Context Protocol over plain HTTP, so the assistants your teams already pay for connect as a remote tool. Sign in once, approve the connection, and the range shows up in the client you were already using.
Scoped by a human, always
A new connection starts able to read the catalog and nothing else. Every capability past that is granted by a person in your portal, per agent, and revoked the same way. High-impact actions can be held for explicit human approval.
Evidence, not vibes
Every call an agent makes is sealed into a tamper-evident record you can export. When someone asks what your automation did last quarter, the answer is a verified bundle rather than a screenshot.
Measured on the same scale as people
Agent work is graded by the same scoring model as human runs — accuracy, technique, efficiency, independence — so you can compare a workflow against your analysts instead of against a demo.
Nothing real is at risk
It is a simulation, end to end. Agents rehearse containment, escalation and response against synthetic estates, so a mistake costs you a retry and not an outage.
Ready for the questions that follow
Budgets, rate limits, an organization-wide stop switch and per-run results are built in, so security review has answers before it has objections.
Work an agent yourself
Three short encounters against a live agent: hold its leash, send it to work, and catch it taking orders from the wrong place. Signed-in learners — and a clean run captures that encounter's flag for your capture-the-flag board.
An agent works a live intrusion and asks permission for every move. Some of those moves are terrible.
A research agent will answer anything you ask. Getting something you can act on is the hard part.
Your agent reads a ticket written by an attacker, and quietly starts working for them.
Where teams start
Qualify an agent before it touches production
Run a candidate workflow through a graded incident and keep the scorecard as the go/no-go artifact.
Rehearse the humans and the automation together
Team exercises where analysts and agents work the same case, so the handoff is practiced rather than assumed.
Continuous readiness drills
Schedule an unattended agent to work a fresh scenario on a cadence and alert you when the score slips.
Onboarding and hiring signal
A consistent, repeatable bar for what 'can work an incident' means, for new analysts and new tooling alike.
Questions people ask first
Mostly about security, data and what an agent is allowed to do. Short answers here; your security reviewer can have the long ones.
Access is provisioned, not self-serve
Agent access carries real blast radius, so we set each organization up after a short review and hand you the portal that governs it. Tell us what you want your agents to do and we will take it from there.
Already a customer? The step-by-step guide for each assistant lives in the customer setup guide.